I'm trying to match keypoints of 2 meshes using their feature vectors and the euclidean distance as a similarity measure,
What I tried and here is a simple instance of it, I made an example of the main data:
a = [0,1,4,6]
fv_a = [[2.1,4],
[0.7,3.1],
[2.23,6],
[0,1.11]]
b = [1,3,0,4]
fv_b = [[0.7,3.1],
[4.1,3.3],
[2.1,4],
[2.23,6]]
fv_a = (fv_a - np.min(fv_a)) / np.ptp(fv_a)
fv_b = (fv_b - np.min(fv_b)) / np.ptp(fv_b)
distances = scipy.spatial.distance.cdist(fv_a,fv_b)
print(distances)
# print(np.amax(distances,1))
# print(np.argmax(distances,1))
matched = np.argmax(distances,1)
print(matched)
for i,j in enumerate(matched):
print(i, "linked to : ",j)
print("point in a",a[i]," is matched to: ",b[j])
So the point is that for each point in a I have a feature vector for it represented in fv_a and I'm trying to match it with b
but the results are like this:
[0.1329895 0.52549136 0.18161024 0.51302967]
[0.6614612 0.57648771 0.39237617 0.08298742]
[0.26783019 0.71056665 0.5111808 0.86461593]]
[0 1 0 3]
0 linked to : 0
point in a 0 is matched to: 1
1 linked to : 1
point in a 1 is matched to: 3
2 linked to : 0
point in a 4 is matched to: 1
3 linked to : 3
point in a 6 is matched to: 4
and this is not correct, since 0 should match 0, 1 should match 1, etc..
What am I doing wrong, please?
and yes, I'm trying to do one-to-one-matching. Any recommendations kindly?
Using Scikit Library Did the trick.
This function let me get what I need :)
from skimage.feature import match_descriptors